Recently, various input methods for controlling electronic products have been introduced. Methods such as voice recognition, motion recognition, etc. are being developed as such methods for controlling electronic products. Such methods can be used to remotely control products, and are being spotlighted as smart input methods suitable for the “smart” era since a command is transferred using a portion of a human body.
When a conventional remote controller is replaced by the motion recognition, since the remote controller is not used, it is not necessary to search for a remote controller and manage various types of remote controllers. That is, it is not necessary to search for the remote controller, and a target device can be controlled by a motion of a hand on the spot by replacing a basic operation of the remote controller with a predetermined motion suitable for the target device to be controlled. Accordingly, the technology can create a large market due to its convenience.
Meanwhile, it is necessary to process an image for recognizing a motion of a human in order to recognize the motion. When processing the image, the most important issue is to separate the human's body from the background.
A method capable of solving this problem is to use a three-dimensional depth camera. That is, since brightness with respect to a distance is represented on an image when using the three-dimensional depth camera, an object located in the foreground is represented more brightly, and thus the human's body and the background may be separated by extracting relatively brighter portions using this feature, and the human's motion can be found by modeling each joint. However, since the three-dimensional depth camera is very expensive, there is a limitation in replacing low-priced remote controllers.
Accordingly, studies on replacing remote control functions by recognizing hands using a two-dimensional camera have been made, and many patent applications including Korea Patent Application Publication No. 10-2011-0033318 have been filed. However, most of these methods include an operation of recognizing a motion of a hand by extracting a hand region.
However, there are many problems when recognizing the hand from the image. That is, separating the hand and the background is not a simple task, and requires complex image processing having a high error rate. Further, when skin color is used in order to separate the hand, it is not easy to recognize a human having a different skin color, and it is not stably operable according to a change in lighting conditions, etc. When a hand is unable to be recognized through extraction, it is difficult to proceed with a hand recognition operation.